1 kyu difference means 1 handicap stone for an even game. 3 kyu difference is 3 handicap stones for an even game. It is not transitive or linear. 1 handicap stone means no komi, so the 1 stone difference is half as big as the step between 2 stones and 3 stones. Larger handicap give larger advantages. 9 stones handicap is fair for something like 12 ranks apart. Games with large handicaps require different strategy, playing tricky moves rather than sound moves, so programs that play well evenly won't play so well at large handicaps.
There is more ELO difference for one handicap stone among stronger players (since there is less variance in their play). I think it would be best to continue to use ELO, since it is really hard to calibrate the kyu scale without playing handicap games. In human play it is traditional to change the handicap after 3 consecutive wins against the same player, one one person's rank is measured against the pool of people at the same rank. David > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of Don Dailey > Sent: Tuesday, December 12, 2006 8:54 PM > To: [EMAIL PROTECTED] > Cc: computer-go > Subject: Re: [spam probable] [computer-go] Anchor Player > > > Hi Sylvain, > > I'm not worried about the ELO situation but you are right. > When skill is measured by ELO you are talking about the > probability of winning a game against any given opponent, we > just have to be careful how we interpret or compare to other > board sizes. > > If 2 opponents are 50 ELO rating points apart in 9x9, they > will be much farther apart in the 19x19 game assuming their > absolute strength (whatever that means) stays the same at > other board sizes. > > The idea of changing the server to accommodate the kyu system > intrigues me. I might need a little help understanding this > however since I don't fully understand the kyu system. > > Does a 1 kyu difference mean I can give you 1 stone if I am > better and expect to come out about even? > > Does this all work out in a transitive way? If a 6 kyu can > give a 7 kyu 1 stone, and the 7 kyu can give an 8 kyu 1 > stone, can the 6 kyu expect to play even with the 8 kyu > player giving 2 stones? > > Of course internally I would use fractions of a kyu or even > ELO and convert. I would probably need some help determining > the right formula. > > At any rate, I am not willing yet to change CGOS to > accommodate this, I would have to add code to get the rating > system working correctly as well as code for setting up > handicap games. If the 19x19 CGOS becomes popular I would > eventually change it over - especially since it would allow > pairing with handicaps to make all matches roughly equal. > > Would this simple system work: > > 1. Start all players out at the same kyu rating. > > 2. Pair randomly. > > 3. If you win your match, modify kyu rating slightly down. > > 4. If you lose your match, slighly change kyu upward. > > > All this is applied on top of handicaps of course. > > But unless 2 players are an integer kyu apart, a handicap > would be slighly unfair to one side or the other. Is it > sufficient to modify the ratings in linear proportion to the > amount of "unfairness?" > > > - Don > > > > > On Wed, 2006-12-13 at 04:48 +0100, [EMAIL PROTECTED] wrote: > > > GnuGo is another possibility and has the advantage of > being a well > > > known quantity, but Gnugo fails to meet some of the > criteria above > > > such as being too deterministic and using heavy resources. > > > > Hello, > > > > GnuGo at level 0 met almost all requirement I think. Perhaps too > > deterministic, but I even not sure. It is already > relatively strong and plays > > very quickly. > > > > By the way, I have concern about the 19x19 no one > mentionned I think. > > In 19x19 > > the probability of winning between two players will be very > often close to 0 > > or 1. As the game is longer, if a player is better then his > probability of > > winning is very big because he can repair his mistakes. > Then the ELO could > > not be sufficent. Am I wrong? Perhaps you can use kyu > rating, setting > > handicaps (and one advantage is that then the pairing > system is simpler). > > > > Sylvain > > > > _______________________________________________ > computer-go mailing list > computer-go@computer-go.org > http://www.computer-go.org/mailman/listinfo/computer-go/ > _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
